Performances of decision feedback equalizers using neural networks under frequency selective fading channels

Abstract

linear and it uses linear algorithm such as RLS(recursive In this paper we evaluate performances of decision feedback equalixers(DFE) using multilayer neural networks under frequency selective fading channels. A novel DFE is proposed. The proposed DFE uses a neural network which carries out unsupervised learning selectively in a tracking mode. The proposed DFE is different from the DFE using the conventional neural network which carries out learning for every data in the tracking mode. The neural network used in the proposed DFE can avoid false learning caused by incorrect teacher signals with setting the appropriate threshold to decide whether the learning should be carried out or not.The fading channel to be considered is frequency selective and its statical characteristics are Rayleigh. Simulation results show that the performance of the DFE using the conventional neural network is superior to that of the conventional DFE and also show that the performance of the proposed DFE is superior to that of the DFE using the conventional neural network.